exalis

Platform & geospatial engineering · Belgium

From point cloud
to production.

I am Alexander Kerckhofs, a staff-level platform and geospatial data engineer. Through Exalis I build the pipelines that turn LiDAR and imagery into data people use, and run the Kubernetes platforms they depend on.

Start a conversation What I do
points
0
crs
EPSG:3812
buildings
LoD2

Services

Platforms that pass the audit, geodata that lines up to the centimetre, and 3D on the web that stays fast at country scale.

01

Platform engineering

Kubernetes on Azure (AKS), GitOps with Flux, infrastructure in OpenTofu, PostgreSQL operators, identity with Keycloak, GPU node pools. Capacity and cost planned, and built to hold up in an ISO 27001 audit, not just a demo.

02

Geospatial data engineering

LiDAR point clouds, aerial and oblique imagery, LoD2 building reconstruction, and pipelines in Python and Dagster. Coordinate systems done right, down to the Belgian datum grid.

03

3D web & performance

CesiumJS and 3D Tiles viewers that stay smooth with large point clouds: tileset structure, level-of-detail hierarchies and render loops profiled down to the cause.

04

Root-cause investigations

Intermittent failures, timeout cascades and performance cliffs that survive the usual fixes, traced through logs, profiles and packet captures until the cause is proven. Then fixed, or handed to the vendor with evidence they cannot wave away.

Selected work

Six years on one production platform. Client names stay private; the numbers come from the project records.

  1. 40min

    33 of 33 production nodes, no reboots

    Kernel CVE response

    Rolled out a mitigation for a kernel privilege-escalation flaw across every production node on the evening it was flagged, then re-verified it after the next node-image upgrade.

  2. 2% to 92%

    valid 3D building solids

    LoD2 building reconstruction

    Benchmarked two open-source reconstructors on 2,773 buildings before switching engines, at a median roof error of 6.5 cm. The pipeline now publishes more than 400 statistical sectors.

  3. 115 to 14ms

    CPU time per orthophoto tile

    Orthophoto tile server

    Rewrote the read path of a production tile service. Warm latency went from 133 to 21 ms and tiles became ten times smaller, with pixel-identical output.

  4. <1 to 8.75GiB

    memory per database pod, injected silently

    A provider-side policy change

    Production stopped scheduling (0 of 36 nodes fit) after the managed-Kubernetes provider quietly enabled an admission policy that padded every database sidecar. Traced and stopped the same morning, with an evidence pack for the vendor.

Approach

  1. Measure first

    Profiles, traces and benchmarks before opinions. The root cause, not the loudest symptom.

  2. Name the gaps

    You get the trade-offs, open risks and decisions in writing, not an idealised diagram.

  3. Leave it owned

    Runbooks, C4 diagrams, decision records and audit artifacts stay in your repository and your knowledge base.

About

Exalis is a one-engineer consultancy. You work directly with the person who designs and builds the system.

I work in production environments under ISO 27001 and NIS2, with the privileged access that comes with them, and I document as I go. My background runs from Kubernetes platform operations to municipal-scale LiDAR and imagery pipelines. When a decision needs weighing, you get a written evaluation: for engineers who want the details and for a board that wants the conclusion.

Building something that has to hold up in production?

Tell me what you are building. You will get an answer, in Dutch or English, from the engineer who would do the work: a fixed-scope investigation, a written evaluation, a project build or a longer embedded contract.

alexander@exalis.be